Peace in our time? Who will win the war games between the Pentagon and Anthropic?

When I was a kid, one of my favorite films was the 1983 thriller War Games which centers on a US military supercomputer designed to simulate nuclear conflict scenarios, but which ends up threatening to start World War III in the process.

I was reminded of this when reading this week’s New Scientist magazine and an article headlined AIs can’t stop recommending nuclear strikes in war game simulations, which ought to chill you to the bone.

King’s College London set three leading large language models – OpenAI’s GPT-5.2, Anthropic’s Claude Sonnet 4, Google’s Gemini 3 Flash – against each other in simulated war games. The scenarios involved intense international standoffs, including border disputes, competition for scarce resources and existential threats to regime survival.

The article reports that the various AI models played 21 games against one another, taking 329 turns in total, and produced around ...